The Chicken Littles of Artificial Intelligence

CNNMoney, citing a PwC report, declared that 38 percent of USA jobs will be lost due to robots and artificial intelligence over the coming 15 years.  Jobs that perform routine, repetitive tasks and are in industries include manufacturing, banking, education, retail and hospitality.  The same warning bells are being rung by The Economist, New York Times, The Guardian and others.   The World Economic Forum cites a “net loss of over 5 million jobs by 2020 in 15 major developed and emerging economies.”

The mainstream media headlines around automation related job loss are akin to Chicken Little’s warning that the sky was falling. The sensationalism overstates reality.

The impression is that job loss due to automation is a recent phenomenon. It’s not. The ATM was created in 1967 and has taken over 30 years to evolve into what we take for granted today. Did it significantly reduce bank teller jobs?  Yes, over a long period of time. But it did not eliminate the position; it evolved.

There are countless examples of how technology has changed economies, professions and industries.  This has been going long before the steam engine and electricity was invented. The pace of technological change today is, however, increasingly faster.  PwC’s 15 years forecast for job loss is challenging to believe when we look at the reality of what it takes to implement artificial intelligence.

Spoiler alert – our ability to forecast the timing of future events with a measurable degree of accuracy isn’t particularly good.

Rurik Bradbury, Head of Research at LivePerson, a mobile and online messaging company, has a different perspective. “The prospect of replacing entire jobs with just technology is unlikely,” shares Bradbury. “There is a lot of confusion about AI with more talk than actual deployment.”

Today’s artificial intelligence technologies are capable of performing tasks at the atomic level. These are very narrowly defined tasks that operate within a clearly defined set of responses.

Based on LivePerson’s customer experience, Bradbury strongly believes that AI can perform, on average, 40 percent of the tasks comprising customer care jobs.  In other words, AI driving the level of unemployment forecasted by CNN within the next 15 to 20 years is unlikely. Even if we just focus on customer care jobs across all industries.

Here’s why.

Jobs are comprised of a multitude of tasks as well as a wide range of problem solving situations that require lateral thinking and complex, emotion-based human interactions. Bringing in AI to a customer support position, for example, requires the job to be broken down into its detailed components.

On average approximately 40 to 50 percent of tasks in a call center are good candidates for automation. These are tasks that a call center agent or manager can trigger – updating your address, for example. The dialog between the AI and the customer is controlled by how the AI application is programmed and closely measured with human oversight.  AI does not run without tight controls in place.  The analytics include sentiment analysis that tells management which AI-conducted customer interactions were positive or negative.  Negative interactions can result in shifting that task back to a human or reprogramming the AI software.

AI doesn’t replace workers; it augments their ability to be more effective and productive. That doesn’t mean that the nature of work will not change, it will. The operative word is augmented – Bradbury calls it “job sharing.”

Routine, data-driven, narrowly defined subtasks will be automated freeing the human worker to engage in higher level, most sophisticated tasks such as creative problem solving, strategic thinking and relationship building.   The latter being things humans are much better suited for.

Based on Bradbury’s research and LivePerson customers’ experience, the rate of AI taking over human tasks is slower than popular media would lead you to believe.

First, to effectively employ AI to drive a positive, productive customer experience requires a clear plan based on gradual automation over time.  Secondly, the current rate of automating tasks is one percent a year. In the case of the 40 percent of call center agent tasks that could be candidates for automation, companies would be extremely hard pressed to achieve that level of automation within ten to fifteen years. So much for predictions.

That doesn’t mean ignore artificial intelligence. Approach it with a solid plan based on best practices.  Here are a few of Bradbury’s suggestions:

  1. Collect a data set of good (read: successful) customer interactions and categorize them, identifying the most frequent interactions.
  2. Pick candidates for automation based on opportunities to improve the interaction. Start with a very small group of interactions to experiment with.
  3. Take a subset of these identified interactions and create a chatbot or AI interface that is specific to the atomic task being automated. The more granular the definition and automation of the task, the higher the success. 70 percent of current AI tasks fail because they are too general.
  4. Put the AI task into production aside a team of call center agents and test. That means collect data, perform A/B testing, and analyze the conversations and their outcomes. Evolve the AI software over time based on the results of the analysis.
  5. As success is realized, automate additional tasks based on the same testing and analysis approach. Set performance thresholds for each AI task. Keep in mind that AI applications work in tandem with employees and need to be orchestrated are part of a company’s ecosystem.

How we look at technology directly influences how we fit it into our lives.  Don’t think of artificial intelligence as a separate project or technology. Think of it as part of a job and measure it accordingly.

Ignore the chicken littles. Leverage AI where it enhances the customer experience and delivers measurable value add. Start small, get granular, and go slow.

Leave a Comment